Multiple Observations of Signals in Noise

Abstract

Five observations were made by human subjects of a signal consisting of pulsed tone, of known frequency, in noise. Both variable noise, i. e., noise that is statistically independent from one presentation to another, and constant noise, i.e., noise that is exactly the same on each of the 5 presentations, were used. With variable noise, the detectability index improves, as predicted, as the square root of the number of observations. The use of constant noise, which results in less improvement, provides an estimate of the portion of the total noise affecting detection that is of internal origin. The results under different levels of external noise indicate that internal noise is proportional to external noise. A 2d set of experiments employed signals whose frequencies were unknown to the observers, and signals comprised of several widely spaced frequencies. The results are discussed in relation to 3 alternative models of the process of frequency analysis.